Data centers in space: is orbital computing real?
Updated October 5, 2026
GPU-powered compute moving to orbit — from first hardware to constellation-scale filings.
The bottom line
Orbital compute has real hardware and real money behind it, but the economics only close with very cheap launch at very large scale.
What is happening
The pitch is genuine: vacuum cooling, uninterrupted solar power, no land use. The arithmetic is the problem — current designs run roughly four times terrestrial cost per watt, radiation degrades hardware nobody can repair, and downlink bandwidth caps data-heavy workloads.
Production workloads are already running in orbit, so the open question is no longer feasibility but which architecture and which launch economics make the business case close at scale.
Latest signals
September 30, 2026Starcloud signed with Firefly Aerospace for a cis-lunar AI inference payload, targeting no earlier than 2028 — the first planned commercial AI workload beyond LEO.
Source: Starcloud
June 1, 2026Los Angeles startup Orbital filed with the FCC for up to 100,000 compute satellites targeting 10 GW of orbital AI processing.
Source: FCC filing
March 1, 2026Starcloud raised $170M at a $1.1B valuation and filed for up to 88,000 orbital data centre satellites.
Source: Starcloud
January 1, 2026SpaceX filed FCC plans for up to 1 million satellites capable of space-based cloud and AI computing.
Source: FCC filing
Open challenges
- Radiation exposure degrades computing hardware — no hardware replacement possible
- Uplink/downlink bandwidth limits data-intensive workloads
- Economics require massive scale and Starship-class cheap launch
- Thermal management for sustained GPU workloads in orbit
- Regulatory uncertainty — FCC filings for thousands of compute satellites are unprecedented
Companies to watch
Starcloud, Axiom Space, OrbitsEdge, Blue Origin, NVIDIA, Crusoe, SpaceX, Google, Planet Labs, Aetherflux, Orbital.